Box CEO Aaron Levie: CIO Advice on Agentic AI and the Enterprise

Why It Matters

Agentic AI will dramatically boost enterprise efficiency, but successful deployment hinges on robust governance, data access controls, and sustained human oversight, reshaping how CEOs and CIOs plan technology investments.

Key Takeaways

• Agentic coding accelerates development cycles, cutting years to months.
• Knowledge‑work agents require data access controls and human verification.
• CEOs must experience AI psychosis to calibrate realistic expectations.
• Ambitious AI pilots reveal limits faster, driving iterative improvements.
• Enterprise rollout will create new roles focused on AI governance.

Summary

Aaron Levie, CEO of Box, framed agentic AI as the defining enterprise technology of 2026, contrasting its rapid adoption in software engineering with the still‑messy rollout in knowledge work. He highlighted that nearly 70% of the Fortune 500 are Box customers, giving him a front‑row seat to the transformation and the challenges of scaling agents beyond code. Levie explained that AI agents can automate the repetitive, low‑value portions of software projects—library upgrades, testing, and feature scaffolding—compressing multi‑year timelines into months and enabling teams to tackle larger roadmaps. By contrast, knowledge‑work agents struggle with data silos, limited user technical skill, and the inability to verify outputs, making security, access control, and human review essential. A memorable quote from Levie described the “AI psychosis” CEOs experience: an initial wave of existential dread followed by pragmatic realization that agents still need human oversight. He urged leaders to push ambitious pilots, accept early failures, and iterate as models improve, noting that over‑ambitious projects often reveal the technology’s edge faster. The implications are clear: enterprises must invest in AI governance, create new roles to manage data permissions and agent supervision, and adopt an iterative, ambitious approach to realize the promised productivity gains while mitigating risk.

Original Description

Box CEO Aaron Levie joins CXOTalk episode 921 to explain why enterprise AI agents stall in production, how they break existing security models, and what CIOs should fix first.